WO2009038908A1 - Mesures étalonnées de concentration d'analyte réalisées dans des mélanges - Google Patents

Mesures étalonnées de concentration d'analyte réalisées dans des mélanges Download PDF

Info

Publication number
WO2009038908A1
WO2009038908A1 PCT/US2008/073045 US2008073045W WO2009038908A1 WO 2009038908 A1 WO2009038908 A1 WO 2009038908A1 US 2008073045 W US2008073045 W US 2008073045W WO 2009038908 A1 WO2009038908 A1 WO 2009038908A1
Authority
WO
WIPO (PCT)
Prior art keywords
data set
target analyte
spectra
spectral data
sample
Prior art date
Application number
PCT/US2008/073045
Other languages
English (en)
Inventor
Jan Lipson
Thomas J. Lenosky
Jeffrey M. Bernhardt
Original Assignee
C8 Medisensors Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by C8 Medisensors Inc. filed Critical C8 Medisensors Inc.
Publication of WO2009038908A1 publication Critical patent/WO2009038908A1/fr

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue
    • A61B5/1455—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue using optical sensors, e.g. spectral photometrical oximeters
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
    • A61B5/0075—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/145—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue
    • A61B5/14532—Measuring characteristics of blood in vivo, e.g. gas concentration, pH value; Measuring characteristics of body fluids or tissues, e.g. interstitial fluid, cerebral tissue for measuring glucose, e.g. by tissue impedance measurement
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/44—Detecting, measuring or recording for evaluating the integumentary system, e.g. skin, hair or nails
    • A61B5/441—Skin evaluation, e.g. for skin disorder diagnosis
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/48—Other medical applications
    • A61B5/4836—Diagnosis combined with treatment in closed-loop systems or methods
    • A61B5/4839—Diagnosis combined with treatment in closed-loop systems or methods combined with drug delivery
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7232—Signal processing specially adapted for physiological signals or for diagnostic purposes involving compression of the physiological signal, e.g. to extend the signal recording period
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
    • G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
    • G01N21/65—Raman scattering

Definitions

  • This invention relates in general to making calibrated measurements of analytes in samples, which are illuminated with electromagnetic radiation, so as to produce a scattered spectrum. Specifically, the invention permits determination of the concentration of an analyte in complex mixtures, wherein there is significant spectral overlap between the analyte and other compounds present in the sample.
  • the field of chemometrics is primarily devoted to mathematical techniques whereby the concentration or presence of a target analyte can be ascertained from data which contains signals from other compounds.
  • a typical problem occurs when the data consist of spectra, and the sample consists of a mixture of compounds, one or more of the spectra of such compounds overlapping with that of the analyte of interest.
  • there are several classes of additional information which can be helpful in isolating the signal of the analyte.
  • the spectrum of the analyte may be measured in advance in a sample which does not contain other compounds.
  • a preparation can be made without the analyte, and the spectrum of this preparation ascertained. Then the analyte can be added to the preparation, and a second spectrum taken. When the first spectrum is subtracted from the second, the resulting spectrum should be that of the analyte. Similarly, the spectra of other compounds which are thought to be present in the sample can sometimes be measured beforehand.
  • the set of samples wherein the concentration of the analyte has been ascertained by an independent method and which are used to create the calibration of the spectroscopic apparatus is called the "training set.”
  • the new sample or samples wherein the concentration of the analyte is unknown is called the "test set.”
  • the spectra of the analyte or of other substances present in the samples may or may not be known.
  • a common means of ascertaining whether the calibration will properly predict the concentration of an analyte in a new sample is called "cross-validation.”
  • concentration of the analyte in all the samples is measured by independent means.
  • the set is then segmented into two subsets. One of the two subsets consists of the training set and the other subset will be the test set.
  • the concentration of the analyte in the test set is predicted by the calibration which is obtained from the training set. These predictions can then be compared with the actual, independently measured concentration of the analyte in the test set.
  • the assignment of samples to either the training or test sets can be permuted in many patterns, hence, the concentration of the analyte in every sample or subset of samples can be predicted by the remaining samples.
  • the loadings P ip are not the spectra of compounds in the mixture.
  • the principle components are not unique in providing a basis for the plane of closest fit to the data, X.
  • the loadings, P can be rotated in the plane of closest fit, so while the plane is unique, the basis vectors that describe it are not. Therefore, since the P's can be rotated to make a decomposition of equally low error, these spectra can have no inherent relation to the actual spectra of the compounds.
  • the spectra of the compounds are not, in general, orthogonal, whereas the P's are orthonormal by construction.
  • the set of data Y, which is being fit is not very complex in the sense that it could be adequately represented mathematically by relatively few parameters.
  • the probability of obtaining spurious correlations with one or more the components extracted in the decompositions of the data is then not negligible.
  • a calibration constructed from over-fitting will have impaired predictive accuracy for new data.
  • the correlation may not be guaranteed to hold under all conditions of interest, while in the former case there is not even the possibility of a correct determination.
  • the difficulty may be viewed in terms of the algorithm being under-constrained, permitting spurious solutions which nevertheless model the data in the training set. Alternatively, we may say that in such circumstances it may be necessary to make more measurements so that the data being fit becomes increasingly complex. The required number of such measurements, may, however, be unpractically large.
  • the concentration of specific analytes can be determined by analysis of the composite spectrum of a mixture, and spectral variances of multiple samples that arise from sources other than the varying concentration of the analyte can be extracted.
  • a method to mitigate the potential hazard of over- fitting for insufficiently complex data is used.
  • the fundamental strategy of the proposed algorithm is to isolate the predominant contribution of the target analyte in the collected spectral data to a single independent variable, and to create other appropriate independent variables which describe the variances in the observed spectra from sources other than the variations of the target analyte concentration.
  • the spectral data consists of a series of spectra obtained on a series of different samples, where the number of distinct spectra obtained on each sample can be different.
  • the concentration of the analyte of interest can change during the process of obtaining the spectra; that is, the concentration changes while taking a series of spectra on any given sample and also changes when moving from one sample to the next.
  • the concentration of the target analyte is measured by a means independent of the spectroscopic apparatus, e.g., by an independent reference instrument, so as to create a suitable training set.
  • the proposed algorithm creates a calibration which depends on both the training and the test data, and which can then be applied to predict the unknown analyte concentration for the test spectral data.
  • An estimate of the concentration of the analyte, Y subject to a scaling factor, is obtained by forming the dot product of the spectral data set X with the spectrum of the analyte, g, which was obtained a priori.
  • the data set X (which contains both training and test spectral data) is then modified to create a new data set, X f , where the analyte spectrum, g, has been projected out of X.
  • a new set, X s is now created from Xf, by taking the mean of a subset of the spectra for each sample.
  • One preferred embodiment would be to choose the subset to be the first n spectra in each sample, but many other choices are also possible.
  • the mean sample spectra are intended to capture the sample to sample variance of the spectra not associated with different concentrations of the analytes since the sample spectra are created from a data set where the analyte spectrum has been projected out.
  • Principle components of X s are then extracted with the scores being T s . These scores, T s , are then projected out of the set X f , resulting in the set X g .
  • X g will contain variance information which is associated neither with the varying concentrations of the analyte, nor with the sample to sample variance not associated with the varying concentrations of the analyte.
  • X g will contain variances associated with changes that occurred during the taking of multiple spectra on each sample, except for those changes associated with the changing concentration of the analyte.
  • the principle components of X g are extracted creating the scores T g . It is then possible to perform a regression against the independent measurements of the analyte concentration, Y, using the independent variables Y, T g , and T s .
  • noise terms are added both to X and Y.
  • the noise term added to X has both a multiplicative scaling factor, and is further multiplied by the analyte spectrum, g.
  • the cross-validated error of the estimation has a distinct minimum for the proper choice of scaling of the noise term added to the data X, and it is shown that the probability of over- fitting is substantially mitigated.
  • Fig. 1 is a block diagram of a system for measuring the spectra of samples, and for making independent measurements of analyte concentration, and wherein a calibration is calculated such that the concentration of the analyte can be computed from the spectra for a new sample where the concentration of the analyte is not known, in accordance with one embodiment.
  • Fig. 2 is a flow chart showing a method for creating the calibration from spectral data and from independent measurements of the analyte concentration, in accordance with one embodiment.
  • Fig. 3 depicts a preferred embodiment in which noise terms are added both to the data and to the independent estimates of the concentration.
  • Fig. 4 is a flow chart of another preferred embodiment in which the independently measured analyte concentration is projected out of the spectral data as opposed to the analyte spectrum.
  • Fig. 5 shows another preferred embodiment where the independent measurements of the analyte concentration are modified on the basis of expected systematic differences between the samples measured by the independent apparatus and those measured by the spectroscopic apparatus.
  • Fig. 6 is a particularly preferred embodiment when the spectra are dominated by substances that interfere with that of the analyte, and wherein a revised analyte spectrum is computed to improve the estimation of the analyte concentration.
  • Fig. 7 is yet another preferred embodiment wherein the spectral data is filtered by including or excluding a multiplicity of spectral windows.
  • Fig. 8a is a method for automatically detecting that the sample presented to the apparatus has changed.
  • Fig. 8b is an alternate scheme which is useful for detecting changes for samples that exhibit substantial photo-bleaching.
  • Fig. 9 is an algorithm which is designed to test whether the calibration algorithm has over-fit the data. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
  • Fig. 1 is a block diagram illustrating a system for measuring the spectra of samples in accordance with an embodiment of the invention.
  • a light source 40 illuminates a sample 60 by means of a turning mirror 25.
  • Light scattered from the sample 50 is created by combinations of elastic and inelastic scattering processes within the sample 60.
  • Some of the scattered light is intercepted by the spectroscopic and/or imaging system 20 which contains a detector that converts the optical input to electronic output.
  • the electronic output is transferred to storage 70, which may be for example a hard disk drive.
  • the algorithm which will perform the calibration also resides on the storage unit 70.
  • the spectra of one or more substances believed to be present in the sample may reside on the storage unit 70.
  • a system for making an independent measurement of the concentration of one or more analytes 35 also uploads its output to the storage unit 70.
  • Random access memory 80 can receive any of the information from the storage unit 70 and CPU 90 can perform operations associated with the algorithm.
  • a user interface 95 allows an operator to initiate the algorithm and to view the results of its output.
  • the apparatus of Fig. 1 is suitable for creating both training and test sets.
  • To create a training set multiple samples 60 are presented sequentially to the apparatus.
  • the samples should contain differing concentration of one or more analytes of interest.
  • One or more optical spectra are taken from each sample presented to the apparatus.
  • the concentration of the one or more analytes of interest in each sample should not change significantly during the course of taking a single spectrum, however, their concentration may change during the course of taking multiple spectra. It is not necessary that the same number of spectra be taken on each sample. Independent measurements of the concentration of one or more analytes must be made.
  • the measurements can be made over extended periods of time, such as hours, days or weeks, when it is known that glucose levels will change appreciably.
  • One or more spectra are taken from a specific area of the subject's skin, which shall be referred to as a site. If it is known that the glucose will be changing it is advantageous to take multiple spectra at each such site. Over the course of the measurements, however, it is advantageous to collect multiple spectra from multiple sites on the skin.
  • the blood glucose is independently measured by another apparatus, such as those wherein a blood sample is extracted from the subject and subjected to chemical analysis. The blood samples should be taken such that they correspond in time to the taking of the spectra within a tolerance of about 5 minutes.
  • a suitable training set is constructed from measurements on at least 3 sites and preferably on at least 6 sites, and where the blood glucose varied by at least 50 mg/dL, and preferably at least 100 mg/dL during the course of the sampling.
  • other numbers of sites and variations of blood glucose can also be used to construct a training set. It is generally advantageous to take multiple spectra on each site. This is particularly so if significant changes in blood glucose are being deliberately produced for the purpose of obtaining the calibration. In this particular example, it is advantageous to perform Raman spectroscopy, although the algorithms described are not specific to this technique.
  • each skin site is considered a separate sample (or possibly multiple samples, as described below).
  • the blood glucose concentrations can be estimated for a new sample or samples, which are referred to as the test set wherein the concentration of glucose in the blood is not known, by applying the calibration to the new spectral data.
  • An algorithm suitable for computing the calibration is presented in Fig. 2. The sequence of operations is not important except as will be noted below. In step, 100, setup parameters are loaded.
  • this is administrative information relating to identity of the sample.
  • the spectra for both the training set and the test set are loaded into the matrix variable, X. If multiple spectra are taken on multiple samples, it is necessary to keep track of which spectra correspond to which sample. Such information can be part of the administrative data loaded in step 100 or can be loaded in step 110.
  • the spectrum of a target analyte g.
  • the independent measurements of the concentration of the analyte whose spectrum was loaded in step 110 are also loaded.
  • the spectral data are compressed to improve computational efficiency and to remove noise.
  • An advantageous means of performing the compression is to extract the principle components and accept only the first N of these components in reconstructing the data, where N is chosen to optimize computational efficiency and the quality of the predictions.
  • Particularly useful is to present the compressed data, X c , in the spectral basis of the loadings of the principle components that are accepted. That can greatly reduce the dimension of the matrix X c in the spectral domain. If a new basis is introduced for the spectral data, the analyte spectrum should be put into the same basis, and the revised spectrum is called g c which is g • P, where the P are the loadings of the principle components of X, as is shown in step 140.
  • the spectral data set X f is then orthogonal to the analyte spectrum.
  • the set X f is now used to create another set X s .
  • the set X s is the mean spectrum of the first several spectra, nsp, taken on each sample. Alternatively, it can be the sum of the first nsp spectra, however it is necessary to be consistent in using the same number of spectra for each sample to construct X s . If only a single spectrum was taken on a sample, the spectrum of that sample in X s is just that single spectrum.
  • the set X s is expected to capture the spectral variance associated with each sample that is not attributable to different concentrations of the target analyte. This is so, because the analyte spectrum has already been projected out of the data that is used to create X 8 .
  • step 200 the principle components of X s are extracted, creating the scores T 8 .
  • step 160 the scores T s are projected out of Xf as follows:
  • X g X f - T s * pinv(T s ) * X f (3) where pinv(T s ) is the Moore -Penrose pseudo-inverse of T 8 .
  • the purpose of projecting out the subspace T s of the scores of X s is to remove the sample to sample variance that is not attributable to changes in the analyte concentration.
  • the set X g should contain only the spectral variances associated with neither the sample to sample variation, nor the variances associated with the differences in the analyte concentration.
  • An example of such variances might be changes which transpired in the samples over the course of taking multiple spectra, provided those changes are not associated with the analyte concentration.
  • Such variations can be induced by the process of measurement as for example photo-bleaching produced by the light source, or the temperature of the sample may be changed by the light source or by other causes. Physical pressure on the sample can also be a source of such variations.
  • step 170 the principle components of X g are extracted creating the scores T g .
  • One advantage of step 160 is that the scores T s will be substantially orthogonal to the scores T g .
  • the estimate Y is expected to contain appreciably all the signal from the analyte, however, it is expected also to contain signal from any other substances whose spectra overlap with that of the analyte.
  • R 8 [ F T g TJ ⁇ Y (5) where the symbol 'V is used to indicate linear regression, using the independent variables preceding the 'V symbol upon the dependent variable following it.
  • the regression of Eq. 5 generates regression coefficients, R g , the first of which is a constant term that is additive in creating the prediction. It can be seen that the orthogonality of the T g 's to the T s 's is advantageous in keeping the regression in Eq. 5 well conditioned.
  • the basic idea behind the regression is that Y is a zeroth order estimator for the analyte concentration, and a linear combination of the T g 's and T s 's is used to form a correction.
  • the set of regression terms is not linearly dependent.
  • the T g 's and T s 's are automatically an orthogonal set based on their method of construction, however the variable Y is not necessarily orthogonal either to the T s 's or the T g 's.
  • the reason the colinearity is not generally severe enough to be detrimental in regression is that Y , which is a zeroth order estimator for the analyte concentration, is not likely to be colinear with any linear combination of the T g 's and T s 's, since they were formed from sets where the analyte spectrum was projected out.
  • cross-validation is performed.
  • Cross-validation is a well-known procedure in which a portion Q of the data is used to generate a calibration from which the analyte concentrations are predicted for the remaining data.
  • Q can be set to different permuted subsets of the data, so that each datum is predicted at some point.
  • An error metric based on the quality of the predictions quantifies the quality of the calibration and is also somewhat indicative of over- fitting or the lack thereof. For further explanation on this point, see Chemometrics Data Analysis for the Laboratory and Chemical Plant, Brereton, R.G.,
  • step 220 a subset of the training set is extracted as a new training set, while the remainder is designated as the test set.
  • the new training set is then used to predict the test set, and residuals are computed with respect to the actual independently measured analyte concentrations of the test set.
  • the prediction, Ypred is given by:
  • Ypred R g (l)+R g (2:n'+l)*[ ⁇ T g TJ (6)
  • R g (2:n'+1) is a vector of regression coefficients beginning with the second coefficient and running to the last coefficient
  • n' is the number of columns in the independent variable matrix [ Y T g T s ]
  • R g (l) is a constant obtained from the regression of step 210.
  • the members of the test set and training set can be permuted until all of the set has been predicted by the remainder. If the residuals are satisfactory, the calibration is obtained in a similar manner using the entire training set for training. To do so, the independent variables
  • X n X c +S*n(t)*g c (8)
  • S is a scalar.
  • the appropriate value of S can be calculated on the basis of minimizing the cross-validated error of prediction.
  • Equations 9, 10, 11, 12, 13, and 14 are implemented in steps 155, 300, 163, 310, 320, and 330 respectively.
  • Yet another alternative consists of using the output of step 155 to create X s in step 190, rather than the output of step 150. If this is done, we must find an alternative way to compute X s in the cross-validation, for now it can be seen that we have used Y to compute Xs, which we are not entitled to know in the cross-validation.
  • the procedure in the cross- validation requires us to derive the scores that are analogous to T s in the regression 220, and we can proceed in the same manner as was done in Eqs. 10, 12, 13, and 14 for finding the scores, T gy , that were analogous to T g .
  • the concentration of the analyte cannot be readily ascertained by independent means from exactly the same volume in the sample from which the spectra is obtained.
  • An example of such a case occurs for noninvasive glucose measurements in human skin by spectroscopy.
  • the method of independent measurement often relies on blood samples but blood is not a dominant constituent of the human skin.
  • the bulk of the glucose in human skin measurable by some spectroscopic methods, is dissolved in the interstitial fluid.
  • the relationship between the glucose in the blood and that in the interstitial fluid is determined by diffusion of glucose across blood vessel walls, and by the uptake of glucose from the interstitial fluid by the living cells of the skin. From the foregoing it can be seen that a calibration error could result from using the unmodified blood glucose concentrations.
  • C is the consumption rate by the tissue of glucose
  • Y is the measured blood glucose as a function of time
  • Yr f is the glucose concentration of glucose in the interstitial fluid as a function of time; and Yd is the volume averaged glucose in the skin as a function of time.
  • the cross-validated predictions that emerge from step 220 can be systematically corrected by inverting the delay model of Eqs. 15, and 16 and applying the inverse to the predictions, as is shown in step 223 of Fig. 5.
  • the inverted delay model is applied to the historical data and to the new prediction. Changes in either direction will be exaggerated, and excursions in either direction will increase in amplitude. It is possible to put reasonable physical limits on the slopes of these changes on the basis of known limitations on the rate of change of the analyte concentration. For example, we can require that the rate of change of a blood glucose prediction be less than 4 mg/dL/min.
  • the algorithm would apply the inverted delay model but default to the maximum rate of change if the inverted delay model predicted a change greater than that which is physically possible.
  • Y d instead of Y for the case of noninvasive glucose testing, if the analyte concentration is projected out of any of the data, as in Eq. 9.
  • substances that have spectra that are partially colinear to the analyte spectrum may contribute substantially to the estimate of Y in Eq. 4. This is particularly likely if the concentrations of substances whose spectra are colinear with the analyte are higher than that of the analyte.
  • the Y of Eq. 19 can be used in preference to the Y of Eq. 4, in step 180 of Figs. 2 or 4.
  • r can be substituted for g c in step 185 of Figs. 3 or 5.
  • the constant Ci in Eq. 18 can be chosen so as to minimize the error of the cross-validated prediction of the analyte concentration. It has been found in practice that this is beneficial.
  • a spectrum consists of signal amplitudes at various wavelengths. It is sometimes the case that the inclusion of a subset of the spectral data, at some subset of wavelengths, will actually give rise to errors in the prediction of the analyte concentration that are larger than if the subset had been excluded. At some wavelengths the analyte may have weak signal compared to that of other substances in the mixture. It is not necessarily disadvantageous to include these wavelengths in the analysis as inclusion may sometimes enable a better estimate of the spectral contribution of other substances which contribution may then be more accurately extracted. It is, however, sometimes the case, that inclusion of some wavelengths is deleterious. A simple example of such a case is when the signal consists of random noise at these wavelengths.
  • step 106 where a loop which permutes the on and off states of each window is initiated in step 106, and the loop is terminated in step 227, where the window choices which minimized the cross- validated prediction error are reported.
  • the sequence can start from a state where all the windows are included, but this is not an essential feature. It also can be repeated several times and executed beginning with windows at smaller wavelengths and proceeding to windows at larger wavelengths or vice versa. Any of the methods illustrated in Figs. 2-6 can be nested within the wavelength window optimization loop. If the computational power available is adequate, the method illustrated in Fig. 7 can examine the case of turning windows on and off two or more at a time in all combinations, which may improve the outcome.
  • the site on which the measurement is made may change because the subject moves. These motions can be either deliberate, in which case the change can be noted by an operator, or it may be inadvertent.
  • the method of Fig. 2 attempts to extract the effect of site to site variance by creating a data set consisting of some number of spectra taken at each distinct site.
  • the scores, T s associated with the principle components are vectors having the dimension of the number of sites. To perform step 160 of Fig. 2, it is necessary that the dimension of the T s 's must be the same as the number of spectra in X f . In step 190 of Fig.
  • a data set X s which consists of only a single spectrum at each site (or equivalently for each sample). It is often advantageous, however, that multiple spectra be taken at each site, hence, the data set Xf often consists of multiple spectra at each site and will therefore have dimension greater than T s .
  • the values of the T s 's computed for each site are duplicated as many times as necessary so that the number of values of each T s , at each site is the same as the number of spectra at each site in X f .
  • the S t 's tell us what portion of the spectral data, X c , is colinear with the spectra, P s .
  • the spectra P s capture the site to site or sample to sample variance, not associated with variations in the analyte concentration, since they are loadings of the principle components of Xs. Therefore, the St's will contain the proper information to ascertain a change in site. In particular, if the site is changed, a discontinuity in the S t 's will likely be observed. [073] Next, the discontinuities in the St's are detected as is shown in step 204 of Fig. 8a. We begin by finding the difference between the neighboring St's:
  • Th is a threshold for the Yes/No decision.
  • step 204 provides a 'yes' decision, then it is further necessary to ascertain whether assignment of the spectral data to an additional site will potentially create an excessively small series of spectra for that prospective additional site.
  • data associated with the previous site assignment began at spectrum #25, and that we now detect a discontinuity at spectrum #32. If we assigned data beginning with spectrum #32 to a new site, we would only have 7 spectra associated with the previous site, which in this example would not meet our requirement for a minimum of 10. Furthermore, it is possible that the prospective assignment will not have enough series.
  • Fig. 8a is a suitable general tool for detecting when a sample has changed significantly due to causes not associated with a change in analyte concentration. Although its function was illustrated by using an example of different sites on biological organisms capable of motion, it is not limited to this case.
  • the set X s in Fig. 2 can be created, where the set X s is representative of the variations of the spectra between samples, and either the analyte spectrum or its concentration has first been projected out of the data from which X s was calculated.
  • Discontinuous changes are detected in step 114, by comparing neighboring averaged fluorescence signals.
  • a new site may be assigned if the discontinuity is of sufficient magnitude as in step 119. If not, the original assignment is accepted as in step 117. Considerations of whether series of adequate length can of course be handled as in steps 207, 208, and 209 of Fig. 8a, if desired.
  • step 420 a perturbation ⁇ Y is added to the analyte concentration Y:
  • the perturbation, ⁇ Y is advantageously chosen to be a sine function.
  • the wavelength of the perturbation is best chosen such that the maximum derivative of ⁇ Y, normalized to the amplitude of the perturbation, is similar in magnitude to the maximum derivative of Y, normalized to the variance of Y.
  • Y pre d2 be the prediction produced by the algorithm when Y is replaced with Y p , the perturbed Y.
  • the method now tests to find whether the error of the new estimate associated with the propagation of the perturbation ⁇ Y through the calibration algorithm, has increased according to expectations.
  • the appropriate test is given in step 430 and is:
  • Test is the threshold for declaring the model over-fit and is advantageously chosen to be much less than the amplitude of ⁇ Y. If the test is true, then the data has not been over- fit as shown in step 440. If the test is false then the data is over-fit as shown in step 450. [079]
  • the proposed test is independent of the algorithm used for calibration, and algorithms different from those proposed in Figs. 2-7 may be used to create the necessary inputs for Eqs. 24 and 26.
  • the regression vector used in calibration depends on the choice of both the training and the test data.
  • the training set would be spectra taken under conditions for which the analyte concentration is known from an independent reference meter.
  • the reference data might be from an invasive meter that measures the glucose concentration using glucose oxidase reactions.
  • Either all of the available reference data would be used in constructing a training set, or the training set would be advantageously chosen to be a particular subset of the available reference data. This subset might be, for example, chosen to remove outliers, chosen to optimize the cross-validation, chosen to minimize the size of the training set, or chosen to best model the observed variance in the training set.
  • the test set could be chosen either as a single measurement X test , or perhaps advantageously chosen to also include a set of measurements X ex tra that represent either sample to sample variation or variation within the same sample as X tes t; as explained in Section 20. Because the regression produced by calibration depends globally on the test set, these choices may be advantageous in some instances.
  • X ex tra may be chosen using, e.g., a classifier scheme or expert system, in order to optimize the quality and reliability of the predictions over the course of a continuous measurement made over a long period of time. Because data on, e.g., a human subject are correlated in time, it may also be advantageous to choose Xextra from among recent prior measurements.
  • Certain aspects of the present invention include process steps and instructions described herein in the form of a method. It should be noted that the process steps and instructions of the present invention could be embodied in software, firmware or hardware, and when embodied in software, could be downloaded to reside on and be operated from different platforms used by real time network operating systems. [086]
  • the present invention also relates to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, or it may comprise a general-purpose computer selectively activated or reconfigured by a computer program stored on a computer readable medium that can be accessed by the computer.
  • Such a computer program may be stored in a computer readable storage medium, such as, but is not limited to, any type of disk including floppy disks, optical disks, CD-ROMs, magnetic- optical disks, read-only memories (ROMs), random access memories (RAMs), EPROMs, EEPROMs, magnetic or optical cards, application specific integrated circuits (ASICs), or any type of media suitable for storing electronic instructions, and each coupled to a computer system bus.
  • the computers referred to in the specification may include a single processor or may be architectures employing multiple processor designs for increased computing capability.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Surgery (AREA)
  • Public Health (AREA)
  • Pathology (AREA)
  • Veterinary Medicine (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Optics & Photonics (AREA)
  • Emergency Medicine (AREA)
  • Dermatology (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medicinal Chemistry (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
  • Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)

Abstract

La concentration des analytes dans un mélange complexe peut être vérifiée par mesure spectroscopique, même si les spectres des substances autres que celui de l'analyte chevauchent celui de l'analyte. Les deux concentrations mesurées indépendamment de l'analyte dans une base d'apprentissage et du spectre de l'analyte sont utilisées. Les variances des données spectrales attribuables à l'analyte sont isolées des variances spectrales issues d'autres causes, telles que les changements de composition associés à différents échantillons qui sont indépendants de l'analyte. Dans le cas spécial de mesures non invasives de la concentration de glucose sur la peau d'organismes biologiques, la moyenne de glucose en volume dans l'échantillon est prédite à partir du glucose dans le sang. L'invention concerne également un test permettant le surapprentissage des données.
PCT/US2008/073045 2007-08-13 2008-08-13 Mesures étalonnées de concentration d'analyte réalisées dans des mélanges WO2009038908A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US95557507P 2007-08-13 2007-08-13
US60/955,575 2007-08-13

Publications (1)

Publication Number Publication Date
WO2009038908A1 true WO2009038908A1 (fr) 2009-03-26

Family

ID=40468259

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2008/073045 WO2009038908A1 (fr) 2007-08-13 2008-08-13 Mesures étalonnées de concentration d'analyte réalisées dans des mélanges

Country Status (2)

Country Link
US (2) US7961312B2 (fr)
WO (1) WO2009038908A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3928102A4 (fr) * 2019-02-22 2022-12-14 Stratuscent Inc. Systèmes et procédés d'apprentissage à travers de multiples unités de détection chimique à l'aide d'une représentation latente réciproque
GB2613032A (en) * 2021-11-23 2023-05-24 Rsp Systems As Calibration method and system

Families Citing this family (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10475529B2 (en) 2011-07-19 2019-11-12 Optiscan Biomedical Corporation Method and apparatus for analyte measurements using calibration sets
GB201000179D0 (en) 2010-01-07 2010-02-24 Rsp Systems As Apparatus for non-invasive in vivo measurement by raman spectroscopy
US8873040B2 (en) * 2012-01-19 2014-10-28 Mustard Tree Instruments, Llc Raman apparatus and method for real time calibration thereof
EP2859313B1 (fr) * 2012-05-31 2021-07-07 Richard Jackson Système et procédé pour la détection de la présence de composantes spectrales dans les spectres de mélange
AU2014261003B2 (en) * 2013-05-02 2019-03-28 Atonarp Inc. Monitor and system for monitoring living organisms
US20180180549A1 (en) * 2014-03-25 2018-06-28 Malvern Instruments Ltd. Raman Spectroscopic Structure Investigation of Proteins Dispersed in a Liquid Phase
US10869623B2 (en) 2014-05-28 2020-12-22 Santec Corporation Non-invasive optical measurement of blood analyte
WO2016054079A1 (fr) 2014-09-29 2016-04-07 Zyomed Corp. Systèmes et procédés pour la détection et la mesure du glucose sanguin du sang et d'autres analytes à l'aide du calcul de collision
US10548520B2 (en) 2015-04-01 2020-02-04 Santec Corporation Non-invasive optical measurement of blood analyte
JP6713149B2 (ja) 2015-06-01 2020-06-24 サンテック株式会社 ２つの波長を合成する光コヒーレンストモグラフィーシステム
US10557792B2 (en) 2015-12-31 2020-02-11 Abb, Inc. Spectral modeling for complex absorption spectrum interpretation
US9554738B1 (en) 2016-03-30 2017-01-31 Zyomed Corp. Spectroscopic tomography systems and methods for noninvasive detection and measurement of analytes using collision computing
US10677580B2 (en) 2016-04-27 2020-06-09 Santec Corporation Optical coherence tomography system using polarization switching
US9993153B2 (en) 2016-07-06 2018-06-12 Santec Corporation Optical coherence tomography system and method with multiple apertures
US10426337B2 (en) 2017-06-01 2019-10-01 Santec Corporation Flow imaging in an optical coherence tomography (OCT) system
US10408600B2 (en) 2017-06-22 2019-09-10 Santec Corporation Optical coherence tomography with a fizeau-type interferometer
US10206567B2 (en) 2017-07-12 2019-02-19 Santec Corporation Dual wavelength resampling system and method
KR102408951B1 (ko) 2017-09-18 2022-06-13 삼성전자주식회사 글루코스 노출량 추정 장치 및 방법과, 글루코스 노출량 추정 모델 생성 장치 및 방법
US10502546B2 (en) 2017-11-07 2019-12-10 Santec Corporation Systems and methods for variable-range fourier domain imaging
CA3089818A1 (fr) 2018-01-29 2019-08-01 Stratuscent Inc. Systeme de detection chimique
US11213200B2 (en) 2018-03-22 2022-01-04 Santec Corporation Topographical imaging using combined sensing inputs
US11067671B2 (en) 2018-04-17 2021-07-20 Santec Corporation LIDAR sensing arrangements
US10838047B2 (en) 2018-04-17 2020-11-17 Santec Corporation Systems and methods for LIDAR scanning of an environment over a sweep of wavelengths
KR102574088B1 (ko) * 2018-08-10 2023-09-04 삼성전자주식회사 분석 물질의 농도 추정 장치 및 방법과, 농도 추정 모델 생성 장치 및 방법
US11627895B2 (en) 2018-08-10 2023-04-18 Samsung Electronics Co., Ltd. Apparatus and method for estimating analyte concentration, and apparatus and method for generating analyte concentration estimation model
US20220128474A1 (en) * 2018-10-23 2022-04-28 Amgen Inc. Automatic calibration and automatic maintenance of raman spectroscopic models for real-time predictions
CN109784553B (zh) * 2018-12-29 2022-12-02 沈阳建筑大学 一种室内pm2.5浓度预估方法
US12082910B2 (en) 2019-02-12 2024-09-10 Medtronic Minimed, Inc. Miniaturized noninvasive glucose sensor and continuous glucose monitoring system
DE102020116094B4 (de) * 2020-06-18 2022-02-10 Carl Zeiss Spectroscopy Gmbh Mehrzahl an baugleichen Spektrometern und Verfahren zu deren Kalibrierung

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6167290A (en) * 1999-02-03 2000-12-26 Bayspec, Inc. Method and apparatus of non-invasive measurement of human/animal blood glucose and other metabolites
US20060063993A1 (en) * 2004-08-09 2006-03-23 Dejin Yu Method and apparatus for non-invasive measurement of blood analytes
US20070049809A1 (en) * 2005-07-22 2007-03-01 Kate Bechtel Intrinsic Raman spectroscopy
US20070060806A1 (en) * 2005-04-27 2007-03-15 Martin Hunter Raman spectroscopy for non-invasive glucose measurements

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5807750A (en) * 1995-05-02 1998-09-15 Air Instruments And Measurements, Inc. Optical substance analyzer and data processor
US5668374A (en) * 1996-05-07 1997-09-16 Core Laboratories N.V. Method for stabilizing near-infrared models and determining their applicability
US6560478B1 (en) * 1998-03-16 2003-05-06 The Research Foundation Of City University Of New York Method and system for examining biological materials using low power CW excitation Raman spectroscopy
ATE227338T1 (de) * 1998-03-18 2002-11-15 Massachusetts Inst Technology Vaskularisierte perfundierte anordnungen für mikrogewebe und mikroorgane
US6330064B1 (en) * 2000-03-13 2001-12-11 Satcon Technology Corporation Doubly-differential interferometer and method for evanescent wave surface detection
US20030148391A1 (en) * 2002-01-24 2003-08-07 Salafsky Joshua S. Method using a nonlinear optical technique for detection of interactions involving a conformational change
US7003337B2 (en) * 2002-04-26 2006-02-21 Vivascan Corporation Non-invasive substance concentration measurement using and optical bridge
US7286222B2 (en) * 2003-07-18 2007-10-23 Chemimage Corporation Sample container and system for a handheld spectrometer and method for using therefor

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6167290A (en) * 1999-02-03 2000-12-26 Bayspec, Inc. Method and apparatus of non-invasive measurement of human/animal blood glucose and other metabolites
US20060063993A1 (en) * 2004-08-09 2006-03-23 Dejin Yu Method and apparatus for non-invasive measurement of blood analytes
US20070060806A1 (en) * 2005-04-27 2007-03-15 Martin Hunter Raman spectroscopy for non-invasive glucose measurements
US20070049809A1 (en) * 2005-07-22 2007-03-01 Kate Bechtel Intrinsic Raman spectroscopy

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3928102A4 (fr) * 2019-02-22 2022-12-14 Stratuscent Inc. Systèmes et procédés d'apprentissage à travers de multiples unités de détection chimique à l'aide d'une représentation latente réciproque
GB2613032A (en) * 2021-11-23 2023-05-24 Rsp Systems As Calibration method and system

Also Published As

Publication number Publication date
US8027033B2 (en) 2011-09-27
US20090079977A1 (en) 2009-03-26
US20110037977A1 (en) 2011-02-17
US7961312B2 (en) 2011-06-14

Similar Documents

Publication Publication Date Title
US8027033B2 (en) Calibrated analyte concentration measurements in mixtures
Heise et al. Noninvasive blood glucose sensors based on near‐infrared spectroscopy
JP6947531B2 (ja) 濃度予測方法、濃度予測プログラム、および濃度予測装置
EP0415401B1 (fr) Méthode et appareil pour la correction de signal multiplicatif
Xu et al. Ensemble preprocessing of near-infrared (NIR) spectra for multivariate calibration
Goodarzi et al. Selection of the most informative near infrared spectroscopy wavebands for continuous glucose monitoring in human serum
US20160100777A1 (en) Intrinsic raman spectroscopy
JP2003524761A (ja) 分光学的な較正モデルを作るための方法及び装置
WO2005103945A2 (fr) Procede et appareil pour effectuer une estimation affinee d'une propriete d'une substance a analyser par transformation de zones multiples
WO2001063251A1 (fr) Procede non invasif d'evaluation de l'epaisseur de la peau et de caracterisation in-vivo des couches de tissus de peau
JP2008530536A (ja) 検体の非侵襲性を測定する方法および装置
Mamouei et al. Comparison of wavelength selection methods for in-vitro estimation of lactate: a new unconstrained, genetic algorithm-based wavelength selection
Suryakala et al. Investigation of goodness of model data fit using PLSR and PCR regression models to determine informative wavelength band in NIR region for non-invasive blood glucose prediction
Garcia-Garcia et al. Determination of biochemical parameters in human serum by near-infrared spectroscopy
JP4329360B2 (ja) グルコース濃度の定量装置
WO2001050948A2 (fr) Procede non invasif pour determiner in vivo l'epaisseur de la peau et caracteriser les couches de tissu cutane
JP2010082246A (ja) 生体スペクトルの測定データ処理方法
KR100545730B1 (ko) 라만 분광법을 이용한 소변 성분 분석 시스템 및 그 방법
CN113974618B (zh) 基于水峰血糖修正的无创血糖测试方法
Idrus et al. Partial least square with Savitzky Golay derivative in predicting blood hemoglobin using near infrared spectrum
JPH11342123A (ja) 非侵襲生体成分測定装置
JP2017060640A (ja) 判定方法、判定装置
Abookasis et al. Application of spectra cross-correlation for Type II outliers screening during multivariate near-infrared spectroscopic analysis of whole blood
JP2004321325A (ja) 血糖値の定量方法
Chauchard et al. Localization of embedded inclusions using detection of fluorescence: Feasibility study based on simulation data, LS-SVM modeling and EPO pre-processing

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 08831480

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 27/05/2010)

122 Ep: pct application non-entry in european phase

Ref document number: 08831480

Country of ref document: EP

Kind code of ref document: A1